Parameter optimization method and device for unified power quality controller based on active disturbance rejection control

Abstract

The application discloses a parameter optimization active disturbance rejection control method and device of a unified power quality controller, and the method comprises the following steps: collecting load side voltage and load side current of the unified power quality controller; inputting the load side voltage and a load side voltage reference value into an optimized active disturbance rejection controller to obtain a series side converter output voltage reference value, which is used for controlling the series side converter output voltage; and adopting an atomic search algorithm to perform parameter optimization on controller parameters, and applying the optimized controller parameters to the optimized active disturbance rejection controller. According to the above technical scheme, the active disturbance rejection control can make the system better adapt to external disturbances; the atomic search algorithm is introduced to perform parameter optimization on the controller parameters, and the controller parameters are efficiently set and have practical operability.

Description

Technical Field

[0001] This invention relates to the field of power electronics technology, and in particular to a parameter optimization active disturbance rejection control method and apparatus for a unified power quality controller. Background Technology

[0002] With the large-scale integration of distributed energy resources into the power grid, significant challenges have been brought to the power quality management of traditional distribution networks. The Unified Power Quality Controller (UPQC), with its ability to simultaneously address issues such as grid voltage distortion, voltage sags, voltage dips, and load harmonic currents, has become a core device for improving grid power quality.

[0003] Currently, PI control, commonly used in industry, has the advantages of simple structure and convenient parameter adjustment. However, for UPQC (Upgraded Dynamic Quality Control), due to the need to cope with complex operating conditions such as grid voltage fluctuations and nonlinear load changes, PI control struggles to quickly eliminate errors when disturbances occur and is easily affected by system parameter perturbations. Linear active disturbance rejection control, on the other hand, has superior disturbance rejection performance and can significantly improve the tracking accuracy of UPQC for harmonic voltages and currents.

[0004] However, applying traditional active disturbance rejection control (ADRC) to UPQC faces challenges such as insufficient high-frequency harmonic mitigation capabilities and difficulties in multi-dimensional parameter tuning. On one hand, existing technologies struggle to eliminate 300Hz high-frequency periodic disturbances without steady-state error when dealing with common high-order harmonics like the 5th and 7th harmonics in the grid background voltage, resulting in persistent residual high-frequency ripple on the load-side voltage. On the other hand, due to the multi-parameter characteristics of ADRCs, especially after improvements and extensions to the control structure, the increased parameter dimensions and coupling make traditional trial-and-error methods insufficient for tuning the parameters of complex systems. Summary of the Invention

[0005] Purpose of the invention: This invention provides a parameter optimization active disturbance rejection control method and apparatus for a unified power quality controller, aiming to solve the problems in the prior art where traditional active disturbance rejection control applied to a unified power quality controller is difficult to eliminate the residual high-frequency ripple on the load side voltage caused by high-order harmonics of the grid voltage, and the difficulty in tuning multiple controller parameters of the active disturbance rejection controller.

[0006] Technical Solution: This invention provides a parameter optimization active disturbance rejection control method for a unified power quality controller, comprising: acquiring the load-side voltage and load-side current of the unified power quality controller; inputting the load-side voltage and load-side voltage reference values ​​into the optimized active disturbance rejection controller to obtain a reference value for the output voltage of the series-side converter, used for controlling the output voltage of the series-side converter; inputting the load-side voltage and load-side current into a harmonic detection extractor to obtain a reference value for the output current of the parallel-side converter, used for controlling the output current of the parallel-side converter; wherein, the optimized active disturbance rejection controller includes a tracking differentiator, a linear state error feedback module, a phase-compensated quasi-proportional resonant controller, and a linear extended state observer; the tracking differentiator calculates a smoothed signal and a corresponding differential signal of the load-side voltage reference value based on the load-side voltage reference value; the linear state error feedback module, based on the smoothed signal and differential signal of the load-side voltage reference value output by the tracking differentiator, the estimated value of the load-side voltage and the estimated value of the differential signal of the load-side voltage output by the linear extended state observer, and combining the proportional gain and differential... The gain is calculated to obtain the linear error feedback control quantity; the phase-compensated quasi-proportional resonant controller, combined with the physical resonance gain, the lead phase compensation angle, and the system control gain coefficient, controls the difference between the load-side voltage reference value and the load-side voltage, and calculates the harmonic compensation quantity; based on the linear error feedback control quantity, the harmonic compensation quantity, the total disturbance estimate output by the linear extended state observer, and the system control gain coefficient, the series-side converter output voltage reference value is calculated; the linear extended state observer, based on the load-side voltage and the series-side converter output voltage reference value, combined with the observer's first gain, second gain, and third gain, calculates the estimated value of the load-side voltage, the estimated value of the load-side voltage differential signal, and the estimated value of the total disturbance; the controller parameters are optimized using an atomic search algorithm, and the optimized controller parameters are applied to the optimized active disturbance rejection controller; the controller parameters include: proportional gain, differential gain, physical resonance gain, lead phase compensation angle, system control gain coefficient, observer's first gain, observer's second gain, and observer's third gain.

[0007] Specifically, the unified power quality controller includes: a three-phase power grid, a series-side converter, a parallel-side converter, and a load. The three-phase power grid and the load are connected through a bus. The series-side converter and the bus are connected to the power grid side through a transformer. The series-side converter and the parallel-side converter are connected. The parallel-side converter and the bus are connected to the load side.

[0008] Specifically, the continuous domain transfer function of the phase-compensated quasi-proportional resonant controller is as follows: G QPR (s)=2K real b0ω c (scosφ-ω0sinφ) / (s 2+2ω c s+ω0 2 ), Among them, G QPR (s) represents the transfer function in the continuous domain, K real b0 represents the physical resonant gain, b0 represents the system control gain coefficient, φ represents the lead phase compensation angle, and ω represents the physical resonant gain. c ω0 and ω0 represent the cutoff frequency and the resonant center angular frequency, respectively, and s represents the Laplace operator.

[0009] Specifically, the calculation formula for the tracking differentiator is as follows: v1'=v2; v2'=-r 2 (v1-v r )-2rv2, Among them, v r v1 and v2 represent the smoothed signal and the corresponding derivative signal of the load-side voltage reference value, respectively. r represents the tracking speed factor, and v1' and v2' represent the rate of change of v1 and v2, respectively.

[0010] Specifically, the calculation formula for the linear state error feedback module is as follows: u0=K p (v1-z1)+K d (v2-z2), Where u0 represents the linear error feedback control quantity, K p and K d Z1 and Z2 represent the proportional gain and differential gain, respectively, and Z1 and Z2 represent the estimated values ​​of the load-side voltage and the differential signal of the load-side voltage, respectively.

[0011] Specifically, the reference value of the output voltage of the series-side converter is calculated using the following formula: u=(u0+u QPR -z3) / b0, Where u represents the reference value of the output voltage of the series-side converter, u QPR z3 represents the harmonic compensation amount, z3 represents the total disturbance estimate, and b0 represents the system control gain coefficient.

[0012] Specifically, the calculation formula for the linearly extended state observer is as follows: z1'=z2-β1(z1-u L z2'=z3-β2(z1-u) L )+ b0u;z3'=-β3(z1-u L ), Where z1', z2', and z3' represent the rates of change of z1, z2, and z3, respectively, and β1, β2, and β3 represent the first, second, and third observer gains, respectively. L This indicates the load-side voltage.

[0013] Specifically, the parameter optimization of the controller parameters using the atomic search algorithm includes: calculating the position and velocity of the d-th controller parameter of the i-th atom; establishing a cost function for the atomic search algorithm based on the steady-state accuracy of the load-side voltage and the total harmonic distortion rate of the load-side voltage; the mass and optimization degree of the atom are negatively correlated with the corresponding cost function value; iterating the atom position based on the atom's velocity, interaction force, and mass, and determining whether the controller parameters corresponding to the iterated atom meet the parameter optimization requirements based on the corresponding cost function value.

[0014] Specifically, the parameter optimization of the controller parameters using the atomic search algorithm includes: generating random numbers following a Lévy distribution as the Lévy step size; using the atomic search algorithm to iterate the atomic positions in the next round to obtain the first-optimized atomic positions; calculating the deviation between the first-optimized atomic positions and the globally optimal atomic positions in the current iteration round; weighting the deviation using the Lévy step size and adding it to the first-optimized atomic positions to obtain the second-optimized atomic positions; the weighting of the deviation using the Lévy step size is also affected by a step size scaling factor; the step size scaling factor is related to the cost function improvement rate in the current iteration round, and the cost function improvement rate is based on the cost of the atomic positions in the current iteration round. The cost function value is calculated based on the improvement of the corresponding atomic position in the previous round. When the improvement rate of the cost function is greater than or equal to the upper limit threshold, the step size scaling factor takes the maximum scaling value. When the improvement rate of the cost function is between the upper and lower limits thresholds, the step size scaling factor is obtained by mapping the improvement rate of the cost function. When the improvement rate of the cost function is less than or equal to the lower limit threshold, the step size scaling factor takes the minimum scaling value or the step size scaling factor of the previous iteration. If the cost function value of the secondary optimized atomic position is lower than that of the primary optimized atomic position, then the secondary optimized atomic position is used to replace the primary optimized atomic position in the next iteration; otherwise, the primary optimized atomic position is retained.

[0015] The present invention also provides a parameter optimization active disturbance rejection control device for a unified power quality controller, used to implement the parameter optimization active disturbance rejection control method for a unified power quality controller provided by any one of the present invention.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: Based on active disturbance rejection control, phase-compensated quasi-proportional resonant control is introduced. Active disturbance rejection control enables the system to better adapt to external disturbances such as grid voltage sags and dips, filter and transformer parameter mismatches. Phase-compensated quasi-proportional resonant control can eliminate steady-state high-frequency residual ripple in the load-side voltage. An atomic search algorithm is introduced to optimize the controller parameters, resulting in highly efficient tuning of controller parameters that are practically operable. Furthermore, the Levy flight step size mechanism is introduced to help the atomic search algorithm effectively escape local optima, improving the reliability of controller parameter tuning. Attached Figure Description

[0017] Figure 1 A schematic diagram of the structure and control process of the unified power quality controller provided by the present invention; Figure 2 A schematic diagram illustrating the principle of optimized linear active disturbance rejection control provided by the present invention; Figure 3 A schematic diagram illustrating the principle of the atomic search algorithm incorporating Lévy flight provided by this invention; Figure 4 A flowchart illustrating the implementation of the atomic search algorithm provided by this invention; Figure 5 Comparative convergence curves of the atomic search algorithms provided by this invention with and without Lévy flight; Figure 6 A comparison diagram of the d-axis load voltage waveforms of traditional active disturbance rejection control and the active disturbance rejection control d-axis load voltage waveforms provided by this invention; Figure 7 The waveforms of voltage sag and dip in the power grid when the control method provided by this invention is applied. Figure 8 The waveform of the load-side voltage when using traditional PI control; Figure 9 A waveform diagram of the load-side voltage for applying the atomic search algorithm and optimized active disturbance rejection control provided by this invention; Figure 10 The waveform diagram of the load-side voltage for applying the Lévy flight-introduced atom search algorithm and optimized active disturbance rejection control provided by this invention. Detailed Implementation

[0018] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0019] See Figure 1 This is a schematic diagram of the structure and control process of the unified power quality controller provided by the present invention.

[0020] In this embodiment of the invention, the unified power quality controller includes: a three-phase power grid, a series-side converter (UPQC series side), a parallel-side converter (UPQC parallel side), a nonlinear load, and a filter circuit. The three-phase power grid and the load are connected through a bus. The series-side converter and the bus are connected to the power grid side through a transformer. The series-side converter and the parallel-side converter are connected. The parallel-side converter and the bus are connected to the load side.

[0021] In this embodiment of the invention, the load-side voltage and load-side current of the unified power quality controller are collected; the load-side voltage and load-side voltage reference values ​​are input to the optimized active disturbance rejection controller to obtain the output voltage reference value of the series-side converter, which is used to control the output voltage of the series-side converter; the load-side voltage and load-side current are input to the harmonic detection extractor to obtain the output current reference value of the parallel-side converter, which is used to control the output current of the parallel-side converter. The filter inductance is L, the filter capacitor is C, and the current flowing out of the capacitor into the primary side of the transformer is i. T .

[0022] For the control of the series-side converter, the acquired load-side voltage u L and the set load-side voltage reference value v r The input is optimized by the active disturbance rejection controller, and the output is the reference value u of the output voltage of the series converter. This control quantity is normalized and input to the PWM generator to generate the PWM drive pulse g1 with the corresponding duty cycle.

[0023] For the control of the parallel-side converter, the traditional existing control method is adopted, specifically: the load-side current i is collected. L and voltage u L After harmonic current detection, the output harmonic current i x , with i x As a reference value for the output current of the parallel-side converter, the output current i of the parallel-side converter cl To control the controlled object, the current control output is sent to the PWM generator to generate a PWM drive pulse g2 with a corresponding duty cycle.

[0024] In practical implementation, a bidirectional DC-DC converter and energy storage device are connected to the DC side during the control process. This provides stable voltage support, decouples the interference of DC-side voltage control on AC-side parameter optimization, and adapts to power management requirements under extreme grid environments.

[0025] In practical implementation, the controlled objects of the unified power quality controller include high-frequency voltage fluctuations in the grid, internal dynamic coupling of the filter, and external current disturbances. In actual engineering, accurately measuring and calculating these factors is extremely difficult. To overcome this challenge, this invention introduces second-order linear active disturbance rejection control.

[0026] See Figure 2 This is a schematic diagram illustrating the principle of the optimized linear active disturbance rejection control provided by the present invention.

[0027] Linear Active Disturbance Rejection Control (LADRC) controllers do not rely on precise system models and possess extremely strong robustness. Through a Linear Extended State Observer (LESO), it unifies the uncertainties within the system model and external environmental disturbances into a single total disturbance, performing real-time estimation and proactive compensation. Compared to traditional PID controllers, LADRC resolves the trade-off between speed and overshoot, achieving superior dynamic response accuracy and stability when facing strong disturbances such as voltage dips and load surges.

[0028] In this embodiment of the invention, the optimized active disturbance rejection controller includes a tracking differentiator, a linear state error feedback module, a phase-compensated quasi-proportional resonant controller, and a linear extended state observer.

[0029] In this embodiment of the invention, the tracking differentiator calculates a smoothed signal of the load-side voltage reference value and a corresponding differential signal based on the load-side voltage reference value.

[0030] In this embodiment of the invention, the tracking differentiator is mainly used for signal smoothing, and the calculation formula is as follows: v1'=v2; v2'=-r 2 (v1-v r )-2rv2, Among them, v r The reference value for the load-side voltage is represented by v1 and v2, which represent the smoothed signal and the corresponding derivative signal of the reference value for the load-side voltage, respectively. r represents the tracking speed factor, and v1' and v2' represent the rate of change (derivative signal) of v1 and v2, respectively.

[0031] In this embodiment of the invention, the linear state error feedback module calculates the linear error feedback control quantity based on the smoothed signal and differential signal of the load-side voltage reference value output by the tracking differentiator, the estimated value of the load-side voltage output by the linear extended state observer and the estimated value of the load-side voltage differential signal, combined with the proportional gain and the differential gain.

[0032] In practical implementation, the Linear State Error Feedback (LSEF) module adopts a linear PD control strategy, which calculates the initial control quantity u0 based on the observed states z1 and z2 output by the linear extended state observer and the command signals v1 and v2 output by the tracking differentiator.

[0033] In this embodiment of the invention, the calculation formula for the linear state error feedback module is as follows: u0=K p (v1-z1)+K d (v2-z2), Where u0 represents the linear error feedback control quantity, K p and K d Z1 and Z2 represent the proportional gain and differential gain, respectively, and Z1 and Z2 represent the estimated values ​​of the load-side voltage and the differential signal of the load-side voltage, respectively.

[0034] In practical implementation, after calculating the initial control quantity u0 through the above linear state error feedback, in order to eliminate the impact of unmodeled dynamics such as grid voltage fluctuations, load changes and system internal parameter perturbations on control performance during actual operation, it is necessary to further perform a compensation step for the total disturbance inside and outside the series-side converter.

[0035] In practical implementation, for the design of composite control strategies for converters, it is considered that although traditional linear active disturbance rejection controllers have extremely strong dynamic disturbance rejection capabilities, in actual digital control systems, they are limited by the bandwidth limit of the observer and are prone to steady-state observation errors for certain sub-high frequency periodic disturbances, resulting in residual high-frequency ripple in the load-side voltage.

[0036] To eliminate the aforementioned specific harmonics, directly connecting a conventional quasi-proportional resonant (QPR) controller in parallel within the linear active disturbance rejection controller (ADC) would result in significant attenuation of the actual physical compensation voltage calculated by the QPR, as the QPR output command would pass through the division operation of the system control gain coefficient b0 of the ADC. Theoretically, to overcome this attenuation and achieve the desired harmonic suppression effect, the resonant gain of the QPR must be significantly increased. However, the inherent calculation and sampling delays of digital control systems will produce severe phase lag at high frequencies (such as at specific harmonics around 300Hz). Blindly increasing the resonant gain at this point can easily trigger high-frequency oscillations or even instability in the system.

[0037] To address this issue, the present invention proposes a composite control architecture that combines a phase-compensated quasi-proportional resonant controller with a linear active disturbance rejection controller in parallel. To address the phase lag problem caused by the inherent delay of digital control systems at high frequencies such as 300Hz, this embodiment preferentially employs a phase-compensated quasi-proportional resonant controller to independently calculate harmonic compensation.

[0038] In this embodiment of the invention, the phase-compensated quasi-proportional resonant controller combines the physical resonance gain, the lead phase compensation angle, and the system control gain coefficient to control the difference between the load-side voltage reference value and the load-side voltage, and calculates the harmonic compensation amount.

[0039] In this embodiment of the invention, the continuous domain transfer function of the phase-compensated quasi-proportional resonant controller is as follows: G QPR (s)=2K real b0ω c (scosφ-ω0sinφ) / (s2 +2ω c s+ω0 2 ), Among them, G QPR (s) represents the transfer function in the continuous domain, K real Let b0 represent the system's desired physical resonant gain, b0 represent the system control gain coefficient, φ represent the lead phase compensation angle, and ω represent the lead phase compensation angle. c ω0 and ω0 represent the cutoff frequency and the resonant center angular frequency, respectively, s represents the Laplace operator, and K r =K real To address the order-of-magnitude mismatch caused by the attenuation of the forward channel 1 / b0, K is therefore... r K is obtained by proportional reconstruction. real b0, K r This represents the internal virtual resonant gain.

[0040] In practical implementation, the difference between the load-side voltage reference value and the load-side voltage is used to control a quasi-proportional resonant controller with phase compensation, which then outputs the harmonic compensation amount u. QPR .

[0041] In this embodiment of the invention, the reference value u of the output voltage of the series-side converter is calculated based on the linear error feedback control quantity, the harmonic compensation quantity, the total disturbance estimate output by the linear extended state observer (for feedforward compensation), and the system control gain coefficient.

[0042] In this embodiment of the invention, the reference value of the output voltage of the series-side converter is calculated using the following formula: u=(u0+u QPR -z3) / b0, Where u represents the reference value of the output voltage of the series-side converter, u QPR z3 represents the harmonic compensation amount, z3 represents the total disturbance estimate, and b0 represents the system control gain coefficient.

[0043] In this embodiment of the invention, the linear extended state observer (part of the active disturbance rejection controller) calculates the estimated value of the load-side voltage, the estimated value of the differential signal of the load-side voltage, and the estimated value of the total disturbance based on the reference values ​​of the load-side voltage and the output voltage of the series converter, combined with the first gain, the second gain, and the third gain of the observer.

[0044] In this embodiment of the invention, the calculation formula for the linearly extended state observer is as follows: z1'=z2-β1(z1-u L z2'=z3-β2(z1-u) L )+ b0u;z3'=-β3(z1-u L ), Where z1', z2', and z3' represent the rates of change of z1, z2, and z3, respectively, and β1, β2, and β3 represent the first, second, and third observer gains, respectively. L This indicates the load-side voltage.

[0045] In practical implementation, the control quantity u of the control object (the reference value of the output voltage of the series-side converter) obtained from the aforementioned calculation, and the collected load-side voltage u L The input is fed into the linear extended state observer, and the outputs are the estimated value z1 of the load side voltage, the estimated value z2 of the differential signal of the load side voltage, and the estimated value z3 of the total disturbance caused by the perturbation of the internal parameters of the series converter and the disturbance caused by the external power grid, which are then fed forward to the corresponding control calculation process.

[0046] In practice, the above process can achieve theoretical control of UPQC. However, the introduction of optimized active disturbance rejection control requires simultaneous tuning of multiple controller parameters, which is difficult. Therefore, an atomic search algorithm is introduced.

[0047] In this embodiment of the invention, the Atom Search Optimization (ASO) algorithm is used to optimize the controller parameters, and the optimized controller parameters are applied to optimize the active disturbance rejection controller; the controller parameters include: proportional gain K. p Differential gain K d Physical resonant gain K real The leading phase compensation angle φ, the system control gain coefficient b0, the first observer gain β1, the second observer gain β2, and the third observer gain β3 are all included.

[0048] In this embodiment of the invention, the position and velocity of the d-th controller parameter of the i-th atom are calculated respectively; a cost function for the atom search algorithm is established based on the steady-state accuracy of the load-side voltage and the total harmonic distortion rate of the load-side voltage; the mass and optimization degree of the i-th atom are negatively correlated with the corresponding cost function value; the atom position is iterated based on the atom's velocity, interaction force, and mass, and the controller parameters corresponding to the iterated atom are determined according to the corresponding cost function value to determine whether they meet the parameter optimization requirements.

[0049] In practical implementation, the formula for calculating the position of the d-th controller parameter of the i-th atom in the 0th iteration of the atomic search algorithm is as follows: x i d (0)=lb d +rand(ub d -lb d ), The formula for calculating the velocity of the d-th controller parameter (the d-th of the aforementioned 8 controller parameters) of the i-th atom in the 0th iteration of the atom search algorithm is as follows: v i d (0)=v min d +rand(v max d -v min d ), Where, x i d (0) and v i d (0) represents the position and velocity of the d-th controller parameter of the i-th atom in the 0th iteration of the atom search algorithm, respectively. d and lb d v represents the upper and lower limits of the position of the d-th controller parameter, respectively. max d and v min d represents the upper and lower limits of the speed of the d-th controller parameter, respectively, and rand represents a uniformly distributed random number between [0,1].

[0050] In practice, during each iteration of the atom search algorithm, the controller parameters represented by each atom are substituted into the system model for simulation. The load-side voltage is collected in real time, and the result value of the cost function is calculated. This is to comprehensively measure the accuracy and speed of the control system in the transient response process, as well as its voltage harmonic suppression capability in steady state.

[0051] In this embodiment of the invention, a cost function for the atomic search algorithm is established based on the steady-state accuracy of the load-side voltage and the total harmonic distortion rate of the load-side voltage. The specific calculation formula is as follows: Cost=λ1J IT +λ2J THD +J P , Among them, J IT λ1 and λ2 represent the steady-state accuracy of the load-side voltage and their respective weights, J THD λ1 and λ2 represent the total harmonic distortion rate of the load-side voltage and its weight, respectively. P This indicates the load-side voltage penalty term.

[0052] J IT =∫t c |e(t)|dt, Among them, t c Let e(t) represent time, and let e(t) represent the error between the load-side voltage and the reference load-side voltage at the t-th iteration of the atomic search algorithm.

[0053] J THD =(Σ h∈H V h 2 ) 1 / 2 / V1, Among them, V h V1 represents the fundamental component in the output voltage of the series-side converter, h represents the harmonic order (starting from 2), and H represents the maximum harmonic order.

[0054] In V min <V target At that time, J P =α jp (V target -V min ) 2 ; in V min ≥V target At that time, J P =0, Among them, V target This indicates the minimum threshold voltage on the load side, V. min The actual minimum operating value of the load-side voltage, α jp J represents P The weights are based on the penalty term. When the load-side voltage exceeds the preset range, the value of the cost function can be increased to assist the optimization process.

[0055] In practice, atoms with lower cost function values ​​tend to have larger masses (atomic mass and optimization level are negatively correlated with the corresponding cost function value), and better fitness. Physically, massive atoms can be considered slow-moving with high inertia, acting as gravitational sources. Conversely, small atoms are understood to move rapidly and are easily attracted by other atoms. Therefore, the mass of an atom can be calculated based on its cost function value.

[0056] The mass M of the i-th atom in the t-th iteration i The formula for calculating (t) is as follows: M i (t)=e^-(Fit i (t)-Fit best (t)) / (Fit worst (t)-Fit best (t)), Where e represents the natural logarithm, Fit i (t) represents the cost function value of the i-th atom in the t-th iteration, Fit worst (t) and Fit best(t) represents the cost function values ​​of the worst-performing atom and the best-performing atom in the t-th iteration, respectively.

[0057] For M i Normalize (t) to obtain m i (t).

[0058] In practical implementation, in order to simulate the motion of atoms in molecular dynamics, the interaction forces between atoms are analyzed, and the total interaction force F on the position of the d-th controller parameter of the i-th atom is defined. i d (t) is: F i d (t)=Σ j∈Kbest (rand j F ij d (t)), Among them, rand j F represents random weights. ij d (t) represents the force exerted by the j-th atom on the i-th atom in the t-th iteration, and Kbest represents the set of the K elite atoms with the lowest cost function values ​​in the current iteration. Only the elite atoms are allowed to exert forces on the current atom to accelerate convergence.

[0059] The number of elite atoms K gradually decreases with the number of iterations. In the t-th iteration, the number of atoms K(t) in the elite atom set is: K(t) = N - (N-2)(t / T) 1 / 2 , Where N represents the total number of atoms, t represents the current iteration number, and T represents the maximum iteration number.

[0060] The force F exerted by the j-th atom on the i-th atom in the t-th iteration ij d (t): F ij d (t)=-η(t)(2h ij (t) -13 -h ij (t) -7 )((x j d (t)-x i d (t)) / ||X i (t)-X j (t)||2), Where η(t) represents the gravitational depth weighting function, h ij(t) represents the proportional distance between the i-th and j-th atoms in the t-th iteration, x i d (t) and x j d (t) represents the position of the d-th controller parameter of the i-th atom and the j-th atom in the t-th iteration, respectively. i (t) and X j (t) represents the spatial position vectors of the i-th and j-th atoms in the t-th iteration, respectively, and ||.||2 represents the L2 norm.

[0061] The gravitational depth weighting function η(t) decays exponentially with time, controlling the exploration step size. Its specific expression is as follows: η(t)=α η (1-(t-1) / T) 3 e^(-20t / T), Where, α η The depth weight constant is α. Since the eight parameters of the optimized active disturbance rejection control are of a large order of magnitude, a large α can be defined at the beginning of the iteration to accommodate the optimization of the eight parameters in the optimized active disturbance rejection control. η When a precise search is required, a smaller α is defined. η .

[0062] Wherein, the proportional distance h ij (t) represents the ratio of the Euclidean distance between the two atoms to the collision diameter, and its specific expression is as follows: h ij (t)=(||X i (t)-X j (t)||2) / σ i (t), Where, σ i (t) represents the collision diameter.

[0063] σ i (t)=||X i (t)-(Σ j∈Kbest X j (t)) / K(t)||2+o', Where o' represents a very small constant.

[0064] In practical implementation, in addition to the interatomic interaction forces, it is also necessary to calculate the system geometric constraint forces on the atoms. The purpose is to guide the atoms in the population to the global optimal solution region. Define the geometric constraint force on the position of the d-th controller parameter of the i-th atom, i.e., the force G forcibly pulled by the globally optimal atom. i d (t), the specific expression is as follows: Gi d (t)=λ(t)(x best d (t)-x i d (t)), Where λ(t) represents the Lagrange multiplier, which decays with the number of iterations, x best d (t) represents the position of the d-th controller parameter of the globally optimal atom in the t-th iteration.

[0065] Based on the interaction forces of atoms, their mass, and the force exerted by the globally optimal atom in the current iteration, calculate the acceleration 'a' of the position change of the d-th controller parameter of the i-th atom in the t-th iteration. i d (t), the specific expression is as follows: a i d (t)=(F i d (t)+G i d (t)) / m i (t).

[0066] Based on acceleration a i d Given (t) and the velocity of the atom, iterate over the atom's position, as shown in the following expression: v i d (t+1)=rand(v i d (t))+a i d (t); x i d (t+1)=v i d (t+1)+x i d (t), Among them, v i d (t+1) and x i d (t+1) represents the velocity and position of the d-th controller parameter of the i-th atom in the (t+1)-th iteration, respectively.

[0067] In this embodiment of the invention, the controller parameters corresponding to the iterated atoms are determined based on the corresponding cost function values ​​to determine whether they meet the parameter optimization requirements.

[0068] In specific implementation, the atomic search algorithm iteration (including the atomic search algorithm with the introduction of Levy flight) in this invention can end when the cost function value is lower than a set threshold, or it can end when the number of iterations reaches a set maximum number of iterations T.

[0069] In practical implementation, the atom search algorithm mainly relies on interaction forces and global optimum constraints for optimization. While this method improves the algorithm's optimization ability, it is prone to getting trapped in local optima, leading to premature convergence. Introducing Lévy flight leverages the non-Gaussian random walk property to randomly introduce long-distance jumps within local Brownian motion, giving atoms the ability to mutate. This method significantly improves the algorithm's ability to escape local optima, achieving a dynamic balance between global and local search, and greatly improving the algorithm's optimization accuracy in complex spaces. The principle of this algorithm is as follows: Figure 3 As shown.

[0070] Combination Figure 3 The diagram illustrates the algorithm's principle. Each circle represents an atom. Red atoms represent the optimal atom in the current iteration, green represent elite atoms, blue represent new atoms that can escape local optima, and black and gray represent the remaining atoms. To simplify the image, the forces associated with black atoms are not shown. Red arrows represent the attractive forces from the corresponding atoms, blue arrows represent the repulsive forces, black dashed lines represent the leadership ability of the optimal atom (i.e., the force forcibly pulled by the globally optimal atom), and blue dashed lines represent Levi's flight to escape local optima. In the optimization process of traditional atom search algorithms, individuals in the atom population (taking six atoms X1 to X6, each representing only a one-dimensional parameter) update their positions based on mutual attraction and repulsion. As shown in region 1 of the diagram, which is prone to getting trapped in local optima, atom X1 is subjected to real-time interaction forces from its neighboring atoms, such as F exerted by X5. 15 F applied by X3 13 and the F applied by X4 14As the algorithm iterates into its later stages, the diversity of the atom population decreases, with a large number of atoms clustered within a narrow local optimum region. At this point, the inter-atomic forces acting on atom X1 cancel each other out and tend towards a physical dynamic equilibrium, causing it to only undergo small Brownian oscillations within this local space and lose its ability to explore a better parameter space. To escape local optima, this invention introduces a Lévy flight mechanism during the atom position iteration process. As shown by the blue dashed arrow in the figure, the non-Gaussian random walk characteristic of Lévy flight can occasionally produce long-distance abrupt jumps. After jumping to region 2, X1's position will shift to NEW_X1, escaping the original local optimum trap. In the next iteration, the inter-atomic forces will change, and the remaining atoms will move towards the new optimal atom NEW_X1. This physical displacement based on Lévy flight effectively achieves global exploration of the complex multi-dimensional control parameter space, improving optimization accuracy.

[0071] In this embodiment of the invention, a random number following a Levy distribution is generated as the Levy step size.

[0072] In practice, the Lévy step size s is typically generated using the following formula. levy : s levy =u levy / |v levy |^(1 / β levy ), Among them, u levy It indicates that it follows N(0,σ) u 2 Distribution (mean 0, variance σ) u 2 A randomly generated value v (from a normal distribution). levy Let β represent a randomly generated value that follows an N(0,1) distribution (a normal distribution with a mean of 0 and a variance of 1). levy This represents the stability index, which determines the shape of the distribution tail, and is usually taken as 1.5.

[0073] In this invention, an atomic search algorithm is used to iterate the atomic position in the next round to obtain an optimized atomic position x. i d (t+1); Calculate the deviation between the first-optimized atom position and the global optimal atom position in the current iteration round, use the Levy step size to weight the deviation, and add it to the first-optimized atom position to obtain the second-optimized atom position.

[0074] In practice, the specific expression for the secondary optimization of atomic positions is as follows: x levy,i d(t+1)=x ASO,i d (t+1)+α levy (t)s levy,i (t)(x ASO,i d (t+1)-x best d (t)), Where, x levy,i d (t+1) represents the position of the d-th controller parameter of the i-th atom obtained in the (t+1)-th iteration of the Lévy flight (secondary optimized atom position), x ASO,i d (t+1) represents the position of the d-th controller parameter of the i-th atom obtained in the (t+1)-th iteration of the atom search algorithm (first optimization of atom position), α levy (t) represents the step size scaling factor for the t-th iteration, s levy,i (t) represents the Lévy step size of the i-th atom in the t-th iteration, x best d (t) represents the position of the d-th controller parameter of the globally optimal atom in the t-th iteration.

[0075] In practice, the global optimal atomic position in the current iteration is the one with the lowest cost function value in the atomic population.

[0076] In this embodiment of the invention, the Levy step size is weighted by the deviation, and the weight is also affected by the step size scaling factor α. levy The role of (t): The step size scaling factor is related to the cost function improvement rate Cost'(t) of the current iteration round. The cost function improvement rate is calculated based on the cost function value of the atomic position in the current iteration round compared to the improvement degree of the corresponding atomic position in the previous round. When the cost function improvement rate is greater than or equal to the upper limit threshold γ, high At that time, the step scaling factor takes the maximum scaling value α. levy,max The improvement rate of the cost function is at the upper limit threshold γ. high and lower limit threshold γ low When the step size scaling factor is between these values, it is obtained by mapping the improvement rate of the cost function. When the improvement rate of the cost function is less than or equal to the lower limit threshold of the improvement rate, the step size scaling factor takes the minimum scaling value α. levy,min Or the step size scaling factor α from the previous iteration. levy (t-1).

[0077] Step scaling factor α levy The specific expression for (t) is as follows: When Cost'(t)≥γ high At that time, α levy(t)=α levy,max , In γ low <Cost'(t)<γ high α levy (t)=α levy,min +((Cost'(t)-γ low ) / (γ high -γ low ))(α levy,max -α levy,min ), When Cost'(t)≤γ low At that time, α levy (t)=max(λ levy α levy (t-1),α levy,min ), Where, λ levy This represents the Lévy attenuation coefficient, and max(.,.) represents the function that takes the maximum of the two values ​​within the parentheses, γ. high and γ low These represent the upper and lower thresholds of the improvement rate, respectively, α levy,max and α levy,min These represent the maximum and minimum scaling values, respectively.

[0078] The specific expression for the improvement rate Cost'(t) of the cost function is as follows: Cost'(t)=(Cost(t-1)-Cost(t)) / Cost(t-1), Where Cost(t-1) and Cost(t) represent the cost function values ​​corresponding to the atomic positions in iterations t-1 and t, respectively.

[0079] In practice, the atomic positions in the t iterations can be either the atomic positions obtained by introducing the Lévy flight or the atomic positions obtained by applying only the atomic search algorithm, depending on whether the atomic positions obtained by introducing the Lévy flight are retained in the t iterations.

[0080] In practical implementation, due to the large range in physical meaning and numerical magnitude of the eight controller parameters to be optimized in the active disturbance rejection controller, traditional step size generation methods are prone to causing parameter optimization divergence or convergence stagnation. Therefore, this invention introduces an adaptive step size scaling factor based on the cost function improvement rate, whose dynamic evolution process is defined in the following three stages. The first stage is the global acceleration search period (when the cost function improvement rate is greater than or equal to the upper threshold of the improvement rate). When the cost function improvement rate is high, it indicates that the algorithm has captured a significant descent gradient or successfully crossed a local extremum region through the Lévy flight mechanism. At this time, maintaining the maximum search step size aims to improve convergence efficiency and ensure that the algorithm quickly approaches the optimal solution direction under unconstrained conditions. The second stage is the adaptive transition period (when the cost function improvement rate is between the upper and lower thresholds of the improvement rate). As the optimization process deepens, the cost function improvement rate gradually slows down, indicating that the search trajectory has shifted from global exploration to the target neighborhood. By establishing a proportional mapping relationship between step size and improvement rate, the smoothness of parameter updates can be effectively guaranteed, suppressing the jump phenomenon caused by excessively large step sizes, thereby achieving a smooth switch between global search and local development. The third stage is the local fine-tuning optimization period (when the improvement rate of the cost function is less than or equal to the lower limit threshold of the improvement rate). When the improvement rate of the cost function is lower than the preset threshold, the algorithm enters the fine-tuning stage. This stage aims to fine-tune the neighborhood of the optimal solution to deal with the convergence bottleneck and ensure that the final parameters meet the requirements of high stability control.

[0081] In this embodiment of the invention, if the cost function value of the second-order optimized atomic position is lower than that of the first-order optimized atomic position, then the second-order optimized atomic position is used to replace the first-order optimized atomic position in the next iteration; otherwise, the first-order optimized atomic position is retained.

[0082] In practice, due to the high randomness of the Levy flight, in order to prevent it from flying randomly and causing the results to deteriorate, the optimization degree of the Levy flight atomic positions is determined based on the value of the cost function of the atomic positions obtained by the Levy flight. Only the Levy flight atomic positions with higher optimization degree (secondary optimized atomic positions) are retained to replace the atomic positions obtained by only applying the atomic search algorithm (primary optimized atomic positions).

[0083] See Figure 4 This is a flowchart illustrating the implementation of the atomic search algorithm provided by the present invention.

[0084] See Figure 5 The curves show the convergence curves of the atomic search algorithms provided by this invention with and without the introduction of Lévy flight.

[0085] Depend on Figure 5As shown, the black curve represents the convergence curve of the cost function (Cost) when only the atomic search algorithm is applied, while the red curve represents the convergence curve of the cost function (Cost) when the atomic search algorithm incorporates Lévy flight. It can be seen that Lévy flight improves the atomic search algorithm in terms of escaping local optima.

[0086] See Figure 6 This is a comparison diagram of the d-axis load voltage waveform of traditional active disturbance rejection control and the d-axis load voltage waveform of active disturbance rejection control provided by the present invention.

[0087] Figure 6 The paper demonstrates the load d-axis voltage under conditions where a 2% fundamental amplitude fifth harmonic and a 5% fundamental amplitude seventh harmonic are injected into the power grid from 0.2s to 1.5s, comparing the conventional active disturbance rejection control (ADRC) with the optimized ADRC provided by this invention. Figure 6 The attached diagram on the left shows the load d-axis voltage under traditional active disturbance rejection control. Figure 6 The attached figure on the right shows the optimized active disturbance rejection control load d-axis voltage provided by the present invention. It can be seen that the control method provided by the present invention has a stronger ability to control power grid harmonics and the d-axis voltage is closer to 311V.

[0088] See Figure 7 It is a waveform diagram of voltage sag and dip in the power grid when the control method provided by the present invention is applied.

[0089] Figure 7 The attached diagram on the left illustrates the scenarios where the grid voltage dips by 20%, rises by 20%, returns to 311V, dips by 15%, and is injected with a 2% fundamental fifth harmonic and a 5% fundamental seventh harmonic. Figure 7 The right-hand diagram shows some details of the left-hand diagram. This demonstrates the control method provided by the present invention's ability to manage external disturbances.

[0090] See Figure 8 The waveform diagram is of the load-side voltage using traditional PI control.

[0091] Figure 8 The attached diagram in the upper left corner shows the waveform of the load-side voltage of the series-side converter under PI control. Figure 8 The attached diagram in the upper right corner illustrates the dynamic process of the load-side voltage waveform. Figure 8 The attached image in the lower right corner is... Figure 8 The attached diagram in the lower left corner shows a partial view of the load-side voltage waveform during steady-state operation after applying PI control. The diagram shows that the load-side voltage recovers after 0.00087 seconds. According to... Figure 8 The attached chart in the lower right corner shows that FFT analysis of the waveform yielded a total harmonic distortion of 2.19%.

[0092] See Figure 9The waveform diagram shows the load-side voltage when the atomic search algorithm (without Lévy flight) and optimized active disturbance rejection control provided by this invention are applied.

[0093] and Figure 8 similar, Figure 9 The attached diagram in the upper left corner shows the waveform of the load-side voltage of the series-side converter under the control method provided by this invention. Figure 9 The attached diagram in the upper right corner illustrates the dynamic process of the load-side voltage waveform. Figure 9 The attached image in the lower right corner is... Figure 9 The attached figure in the lower left corner shows a partial display of the load-side voltage waveform during steady-state operation after applying the control method provided by this invention (without introducing Levy flight). It can be seen from the figure that the load-side voltage recovers after 0.00028s, which is faster than... Figure 8 As shown, according to Figure 9 The attached chart in the lower right corner shows an FFT analysis of the waveform, yielding a total harmonic distortion (THD) of 1.98%, which is lower than... Figure 8 As shown.

[0094] See Figure 10 The waveform diagram shows the load-side voltage when the Lévy flight-introduced atom search algorithm and optimized active disturbance rejection control provided by this invention are applied.

[0095] and Figure 8 similar, Figure 10 The attached diagram in the upper left corner shows the waveform of the load-side voltage of the series-side converter under the control method provided by this invention. Figure 10 The attached diagram in the upper right corner illustrates the dynamic process of the load-side voltage waveform. Figure 10 The attached image in the lower right corner is... Figure 10 The attached figure in the lower left corner shows a partial display of the load-side voltage waveform during steady-state operation after applying the control method provided by this invention (introducing Levy flight). The figure shows that the load-side voltage recovers after 0.00021 seconds, which is faster than... Figure 9 As shown, according to Figure 10 The attached chart in the lower right corner shows an FFT analysis of the waveform, yielding a total harmonic distortion (THD) of 0.54%, which is lower than... Figure 9 As shown.

[0096] The present invention also provides a parameter optimization active disturbance rejection control device for a unified power quality controller, used to implement the parameter optimization active disturbance rejection control method for a unified power quality controller provided by any of the present invention.

Claims

1. A parameter optimization active disturbance rejection control method for a unified power quality controller, characterized in that, include: The system collects the load-side voltage and load-side current from the unified power quality controller; inputs the load-side voltage and load-side voltage reference values ​​into the optimized active disturbance rejection controller to obtain the output voltage reference value of the series-side converter, which is used to control the output voltage of the series-side converter; inputs the load-side voltage and load-side current into the harmonic detection extractor to obtain the output current reference value of the parallel-side converter, which is used to control the output current of the parallel-side converter. The optimized active disturbance rejection controller includes a tracking differentiator, a linear state error feedback module, a phase-compensated quasi-proportional resonant controller, and a linear extended state observer. The tracking differentiator calculates a smoothed signal and a corresponding differential signal of the load-side voltage reference value based on the load-side voltage reference value. The linear state error feedback module calculates the linear error feedback control quantity based on the smoothed signal and differential signal of the load-side voltage reference value output by the tracking differentiator, the estimated value of the load-side voltage output by the linear extended state observer and the estimated value of the load-side voltage differential signal, combined with the proportional gain and differential gain. The phase-compensated quasi-proportional resonant controller, in combination with the physical resonance gain, the lead phase compensation angle, and the system control gain coefficient, controls the difference between the load-side voltage reference value and the load-side voltage, and calculates the harmonic compensation amount. Based on the linear error feedback control quantity, harmonic compensation quantity, total disturbance estimate of the linear extended state observer output, and system control gain coefficient, the reference value of the output voltage of the series-side converter is calculated. The linear extended state observer, based on the reference values ​​of the load-side voltage and the output voltage of the series converter, combined with the first gain, second gain and third gain of the observer, calculates the estimated value of the load-side voltage, the estimated value of the differential signal of the load-side voltage and the estimated value of the total disturbance. An atomic search algorithm is used to optimize the controller parameters, and the optimized controller parameters are applied to optimize the active disturbance rejection controller. The controller parameters include: proportional gain, differential gain, physical resonance gain, lead phase compensation angle, system control gain coefficient, observer first gain, observer second gain and observer third gain.

2. The parameter optimization active disturbance rejection control method for a unified power quality controller according to claim 1, characterized in that, The unified power quality controller includes: a three-phase power grid, a series-side converter, a parallel-side converter, and a load. The three-phase power grid and the load are connected through a bus. The series-side converter and the bus are connected to the power grid side through a transformer. The series-side converter and the parallel-side converter are connected. The parallel-side converter and the bus are connected to the load side.

3. The parameter optimization active disturbance rejection control method for a unified power quality controller according to claim 2, characterized in that, The continuous domain transfer function of the phase-compensated quasi-proportional resonant controller is as follows: G QPR (s)=2K real b0ω c (scosφ-ω0sinφ) / (s 2 +2ω c s+ω0 2 ), Among them, G QPR (s) represents the transfer function in the continuous domain, K real b0 represents the physical resonant gain, b0 represents the system control gain coefficient, φ represents the lead phase compensation angle, and ω represents the physical resonant gain. c ω0 and ω0 represent the cutoff frequency and the resonant center angular frequency, respectively, and s represents the Laplace operator.

4. The parameter optimization active disturbance rejection control method for a unified power quality controller according to claim 2, characterized in that, The calculation formula for the tracking differentiator is as follows: v1'=v2; v2'=-r 2 (v1-v r -2rv2, Among them, v r v1 and v2 represent the smoothed signal and the corresponding derivative signal of the load-side voltage reference value, respectively. r represents the tracking speed factor, and v1' and v2' represent the rate of change of v1 and v2, respectively.

5. The parameter optimization active disturbance rejection control method for a unified power quality controller according to claim 4, characterized in that, The calculation formula for the linear state error feedback module is as follows: u0=K p (v1-z1)+K d (v2-z2), Where u0 represents the linear error feedback control quantity, K p and K d Z1 and Z2 represent the proportional gain and differential gain, respectively, and Z1 and Z2 represent the estimated values ​​of the load-side voltage and the differential signal of the load-side voltage, respectively.

6. The parameter optimization active disturbance rejection control method for a unified power quality controller according to claim 5, characterized in that, The calculation yields the reference value for the output voltage of the series-side converter, using the following formula: u=(u0+u QPR -z3) / b0, Where u represents the reference value of the output voltage of the series-side converter, u QPR z3 represents the harmonic compensation amount, z3 represents the total disturbance estimate, and b0 represents the system control gain coefficient.

7. The parameter optimization active disturbance rejection control method for a unified power quality controller according to claim 6, characterized in that, The calculation formula for the linear extended state observer is as follows: z1'=z2-β1(z1-u L );z2'=z3-β2(z1-u L )+ b0u;z3'=-β3(z1-u L ), Where z1', z2', and z3' represent the rates of change of z1, z2, and z3, respectively, and β1, β2, and β3 represent the first, second, and third observer gains, respectively. L This indicates the load-side voltage.

8. The parameter optimization active disturbance rejection control method for a unified power quality controller according to claim 7, characterized in that, The method of optimizing controller parameters using an atomic search algorithm includes: Calculate the position and velocity of the d-th controller parameter for the i-th atom; establish the cost function of the atom search algorithm based on the steady-state accuracy of the load-side voltage and the total harmonic distortion rate of the load-side voltage; the quality and optimization degree of the atom are negatively correlated with the corresponding cost function value; Based on the velocity, interaction force, and mass of atoms, the atomic positions are iterated, and the controller parameters corresponding to the atoms after iteration are determined according to the corresponding cost function values ​​to see if they meet the parameter optimization requirements.

9. The parameter optimization active disturbance rejection control method for a unified power quality controller according to claim 8, characterized in that, The method of optimizing controller parameters using an atomic search algorithm includes: A random number following a Lévy distribution is generated as the Lévy step size. The atomic search algorithm is used to iterate the atomic position in the next round to obtain the first optimized atomic position. The deviation between the first optimized atomic position and the global optimal atomic position in the current iteration is calculated. The deviation is weighted using the Lévy step size and added to the first optimized atomic position to obtain the second optimized atomic position. The Levy step size is weighted for the deviation, and the weights are also affected by the step size scaling factor. The step size scaling factor is related to the cost function improvement rate of the current iteration. The cost function improvement rate is calculated based on the cost function value of the atomic position in the current iteration and the degree of improvement compared to the corresponding atomic position in the previous iteration. When the cost function improvement rate is greater than or equal to the upper limit threshold, the step size scaling factor takes the maximum scaling value. When the cost function improvement rate is between the upper and lower limits thresholds, the step size scaling factor is obtained by mapping the cost function improvement rate. When the cost function improvement rate is less than or equal to the lower limit threshold, the step size scaling factor takes the minimum scaling value or the step size scaling factor of the previous iteration. If the cost function value of the second-order optimized atomic position is lower than that of the first-order optimized atomic position, then the second-order optimized atomic position is used to replace the first-order optimized atomic position in the next iteration; otherwise, the first-order optimized atomic position is retained.

10. A parameter optimization active disturbance rejection control device for a unified power quality controller, characterized in that, A parameter optimization active disturbance rejection control method for implementing a unified power quality controller according to any one of claims 1 to 9.

Images